Preadaptation in Populations of Neural Networks Evolving in a Changing Environment

Populations of simple artificial organisms modeled as neural networks evolve a preference for one particular food type in an environment that contains more than one food type if the quantity of energy extracted from each food type is allowed to coevolve with the behavioral preference (evolvable fitness formula). If, after the emergence of the food preference, the preferred food gradually disappears from the environment at the evolutionary time scale, the evolved specialist strategy is maintained until the preferred food type has completely disappeared. Then a new specialist strategy suddenly emerges with a preference for another food type present in the environment. The appearance of the new strategy takes very few generations, in fact much fewer than in a population starting from zero (random initial population) in the same environment. This, together with the fact that the population with an evolutionary past is more efficient than the population starting from zero, suggests that the former population is preadapted to the changed environment. An analysis of the activation values of the hidden units indicates that the new food preference can be an exaptation, that is, a new adaptation based on a structure that has previously emerged for adaptively neutral reasons.